Session 4 "Robots & Speech"
Sergej Kozlov (BASNET)
Resources and algorithms of text processing for Belarusian and Russian text-to-speech synthesis
Presenter: Dr. Yury Hetsevich, United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Speech Synthesis and Recognition Laboratory, Belarus
Algorithms of preliminary linguistic text processing and normalization in Belarusian and Russian have been developed. Algorithms for creating and completing electronic grammatical dictionaries have also been developed; this has made it possible to create large (more than
2 million word forms) Belarusian and Russian dictionaries for text-to-speech synthesis in Belarusian and Russian languages. The dictionaries obtained in the international linguistic format NooJ allow linguists to define accentual and morphological structure of a large thematic range of text, and also to cut the physical volume of text files by 5 times for Belarusian and by
8 times for Russian in comparison with known grammatical dictionaries.
Algorithms of syntagmation have been developed based on frequency features of distribution of lexical and grammatical categories of words in texts and based on the use of deep syntactical analysis, which makes it possible to define syntagmas in texts with 45% greater accuracy for Belarusian, and 13 % greater for Russian in comparison with known algorithms for the same test materials.
An experimental set of programs for bilingual text-to-speech synthesizer has been developed based on interaction of general language-independent algorithms of text, prosodic, phonetic and acoustic processors with relevant language-dependent dictionary resources of Belarusian and Russian.
Building metric maps of the mobile robots environment for intelligent identification of objects and processes using Internet distributed computing resources
Presenter: Herasiuta Siarhei , United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Belarus
Co-Authors: Ryhor Prakapovich, Vlad Sychyov
In our presentation we plan to cover the important process of creation metric maps using mobile robots in indoor environment. These maps are not regular floorplans. Floorplans are the one of the step to metric maps. Metric maps are 3D by default and based on high quality laser range finder points arrays and set images from cameras with regular environment additional sensor data (temperature, angles and etc.).
These maps should be created maximum autonomous, and stored in a high performance network infrastructure where they are can be transferred to another network connected robots.